Skip to main content
main-content

Über dieses Buch

Exploration and Innovation in Design is one of the first books to present both conceptual and computational models of processes which have the potential to produce innovative results at early stages of design. Discussed here is the concept of exploration where the system, using computational processes, moves outside predefined available decisions. Sections of this volume discuss areas such as design representation and search, exploration and the emergence of new criteria, and precedent-based adaptation. In addition, the author presents the overall architecture of a design system and shows how the pieces fit together into one coherent system. Concluding chapters of the book discuss relationships of work in design to other research efforts, applications, and future research directions in design. The ideas and processes presented in this volume further our understanding of computational models of design, particularly those that are capable of assisting in the production of non-routine designs, and affirm that we are indeed moving toward a science of design.

Inhaltsverzeichnis

Frontmatter

1. Introduction and Overview

Abstract
Design is among the more complex tasks that humans perform. It is the process of producing artifacts that have desired properties and meet some functional requirements. Designing is a central skill in many human tasks: Architects deal with shape and form to create new buildings, financial planners manipulate money to design profitable portfolios, and mechanical engineers design functional machines with parts such as gears and cams.
D. Navinchandra

2. A Design Scenario

Abstract
In this chapter a hypothetical scenario, taken from the domain of landscape design, traces the steps that might be taken by a machine problem solver as it goes about solving a design problem6. Interspersed with this is an account of how a computer program could be implemented to take similar design steps.
D. Navinchandra

3. The Design Problem

Abstract
Formulating a design task as a computer solvable problem involves two major steps: choosing a representation for the artifact, and choosing a representation for the criteria that will be used to evaluate generated solutions. The formulation we use is based on the classic labeling problem [Waltz 75]. The labeling idea has been extended to suit the design domain by adding the notions of consistency [Nadel 85a] and optimization [Navinchandra 86a].
D. Navinchandra

4. Design Synthesis

Abstract
Synthesis is the process of composing and combining parts to form a whole. It is the process by which design solutions are generated. From a methodological point of view, synthesis can be viewed as a heuristic search process.
D. Navinchandra

5. Design Exploration

Abstract
Exploration is the process of generating and evaluating design alternatives that normally would not be considered. Normal synthesis processes, such as search, are aimed at considering only those design alternatives that are within the solution space defined by the governing criteria. An exploratory process, on the other hand, tries to generate a wide variety of alternatives from outside the solution space, some of which may unearth new opportunities or solve design problems in unexpected ways.
D. Navinchandra

6. Design Adaptation

Abstract
In the previous chapter we discussed how precedent-based knowledge can be used by a program while examining an alternative. This form of reasoning can be used not only for recognizing problems and opportunities, but also for adapting designs that have promise but need a few corrections. Precedents can be used to solve design problems either directly or analogically, as part of a process called design adaptation.
D. Navinchandra

7. Putting It All Together: A Detailed Architecture of CYCLOPS

Abstract
In this chapter we will see how the techniques described in the last four chapters fit into one coherent system. Recall that in Chapter 1 we introduced the program’s architecture in a simplified form (Figure 1—4, Page 14), we will now take a detailed look at it.
D. Navinchandra

8. Relationship to Other Work

Abstract
In this chapter we will compare the techniques used in CYCLOPS to those of other systems. Readers who wish to get a broader sense of the techniques used in Design Automation (DA) systems may refer to the following papers [Mostow 85, Sriram 86, Tong 86, Navinchandra 90].
D. Navinchandra

9. Assumptions, Shortcomings, and Future Research

Abstract
CYCLOPS’ approach makes several assumptions and its current implementation has several shortcomings. In the future, a list of issues will have to be tackled, among them are: Hierarchical approach. CYCLOPS views designs as consistent labeling problems. The representation is flat and and all at one level of detail, making it very difficult to tackle large problems. If a landscape problem which required locating a hundred houses on a large landscape, the combinatorial explosion could grind any computer down to a halt. On the other hand, if we divided the hundred houses into clusters and grouped clusters into neighborhoods, we could design at different levels of abstraction. One of the biggest challenges in developing a hierarchical algorithm for Consistent Labeling Optimization Problems is to find some way of assuring non-dominance of generated solutions; a non-dominated solution at one level of abstraction is not guaranteed to remain non-dominated at lower levels of abstraction.
D. Navinchandra

10. Epilogue

Abstract
In order to innovate, one must have an open mind. One must be willing to relax one’s constraints in order to examone alternatives. Innovation comes from one’s ability to break the rules, to look beyond the norms and to avoid mind-sets. In fact, many of the ideas about constraint relaxation presented in this book are themselves a result of constraint relaxation! As I recall, the hardest task in applying constraint relaxation to a problem is not the relaxation of the constraints but their identification in the first place. We often solve problems making too many assumptions about the world. Our training forces us to impose constraints without being aware of them. For example, consider the following question: ‘Why do flat mirrors flip images left to right and not up to down?’. Many people have problems answering this question, as they make assumptions about mirrors that are really not necessary. Computer models such as CYCLOPS can handle constraints that are explicit in the representation. People, however, can solve problems by extending or changing their view of the problem by switching among representations. We have yet to understand this process.
D. Navinchandra

Backmatter

Weitere Informationen